Childhood Determinants of Internal Youth Migration in Senegal · 2020-02-14 · 5 of international migration, the study of internal migration has been far more limited, partly due

Childhood Determinants of Internal Youth Migration in Senegal

Catalina Herrera Almanza* and David Sahn**

February 2020

* Assistant Professor of Economics and International Affairs, Northeastern University, College of Social Sciences andHumanities, Boston, MA 02115, [email protected]

** International Professor of Economics, Division of Nutritional Sciences and Department of Economics, Cornell University and Director of the Cornell Food and Nutrition Policy Program, Ithaca, NY; Institute for the Study of Labor (IZA), Bonn Germany; Centre d'Etudes et de Recherches sur le Développement International (CERDI), l’Université d’Auvergne, France.

1

Childhood Determinants of Internal Youth Migration in Senegal

BACKGROUND Internal migration, mostly composed of young adults and the poor, constitutes the largest flow of people in developing countries. Few studies document the patterns and determinants of internal youth migration in sub-Saharan Africa. OBJECTIVES This paper analyzes the socioeconomic determinants of the decisions among young adults to internally migrate in Senegal. We focus on whether their decisions to migrate are influenced by individual characteristics, as well as the circumstances in the households and communities where young adults grew up, and whether these factors are differentiated by gender. METHODS Using a unique migration household survey in Senegal, we estimate multinomial logit models to analyze the role of childhood socioeconomic determinants in later youth migration decisions to rural and urban areas. RESULTS We find that young people undertake mostly rural-to-rural and urban-to-urban migrations, and over half of them are temporary migrants. We also find that the determinants are heterogeneous by gender and destination. The higher the fathers’ education, the more (less) likely are their daughters to move to urban (rural) areas. Young individuals who spend their childhood in better-off households are more likely to move to urban areas. The presence of younger siblings during childhood increases the propensity of moving to rural areas. Access to primary schools from the childhood residence decreases the likelihood of migrating to urban areas for both men and women.

CONTRIBUTION We contribute to the sparse literature on internal youth migration in developing countries by highlighting the relevance of the family- and community-level characteristics during childhood in predicting later migration in life. Keywords: Internal migration, youth, Senegal Word Count all submission: Word Count no references and notes:

2

Introduction 1

Internal migration, mostly composed of young adults and people from the lower end of the income 2

distribution, constitutes the largest flow of people in developing countries (UNDP, 2009). 3

Although recent empirical evidence has focused on the analysis of the determinants and impacts 4

of international migration, the study of internal migration has been far more limited, partly due to 5

the lack of reliable data and because it is less politically salient. Few empirical studies have 6

documented the drivers of internal youth migration in developing countries and whether these 7

determinants are differentiated by gender. In this context, family and social factors weigh in the 8

decisions of young adults to migrate. Households face labor and financial market constraints, and 9

migration can be a strategy to diversify income sources and cope with risks, compensating in some 10

cases for the absence of insurance markets (Rosenzweig and Stark, 1989; Stark, 1991; Giles, 11

2007). Families might encourage younger members to migrate, both sons and daughters, not only 12

because they have higher earnings potential in the destination locations, but also because they are 13

more likely to remit money (Taylor, 2001; Heckert, 2015). Furthermore, family and 14

socioeconomic circumstances during childhood can influence the probability of migrating later in 15

life (Abramitzky et al., 2013). 16

This paper analyzes the socioeconomic determinants of the decisions among young adults 17

to internally migrate in Senegal. We focus on whether the decision to migrate is influenced by 18

individual characteristics, as well as the circumstances in the households and communities where 19

young adults grew up, and whether these factors are differentiated by gender. The study of internal 20

youth mobility is particularly pertinent in Senegal, where, like much of sub-Saharan Africa, 64% 21

of the population is less than 25 years old, 59% of the population lives in rural areas, and internal 22

3

migration plays a critical role in the expansion of economic opportunity and social mobility (de 23

Brauw et al., 2014).1 24

More broadly, the analysis of the socioeconomic determinants of internal migration is 25

critical in the context of developing countries, where rural-to-urban migration occurs in 26

conjunction with economic development as rural economies undergo structural transformation 27

(Taylor and Martin, 2001). Although internal migration is widespread in Africa, more than half of 28

the population still lives in rural areas, and given the large and positive income differentials 29

between urban and rural areas, rural-to-urban migration rates might be expected to be even higher 30

in the future (de Brauw et al., 2014). Furthermore, recent studies have highlighted that rural-to-31

rural, and even the reverse urban-to-rural flows have gained traction as internal migration flows in 32

some francophone African countries (Beauchemin and Bocquier, 2004; Beauchemin, 2011). 33

Senegal follows several of these regional patterns of internal migration. Previous research 34

indicates that most of the internal migrants are young people, aged 15 to 34 years old, and the 35

majority of them migrate to look for employment opportunities (Ba et al., 2017). However, family 36

reasons such as marriage are the most important reasons for women’s internal migration (Chort et 37

al., 2017). Most migration gravitates toward urban areas, especially to Dakar, which is not 38

surprising in light of the fact that there are large disparities in education and income between rural 39

and urban areas. For instance, in 2005, the poverty rate was 37% in urban areas and 59% in rural 40

areas. While the average years of education was 7.3 in urban areas, it was only 4.8 in rural areas.2 41

1 According to the 2002 census, a date close to our study, 59% of the population lived in rural areas. More recent figures estimate that this percentage has decreased to 53% (World Bank, 2019). 2 The average years of education is calculated among the population ages 15–19, and the data source is the 2005 Demographic Health Survey. We selected data in 2005 because it is a year close to our survey (2003). Nevertheless, more recent data in 2014 show that the average years of education has increased to 6.5 in rural areas, closing the gap with urban areas. However, the gap in poverty rates between rural and urban areas has been about the same (World Bank, 2019).

4

There is, however, also a considerable amount of rural-to-rural migration from semi-arid regions 42

(Middle Valley of the Senegal River) toward the Groundnut Basin (that is, mainly seasonal 43

migrants working in groundnut cultivation). Furthermore, while small in proportion to other 44

internal migration flows, there is even some urban to rural population movements, mainly in the 45

form of migrants who return to invest in the agricultural sector and who build homes in their 46

villages of origin (Ba et al., 2017). 47

Information on internal migration in sub-Saharan Africa is rare. Except for some recent 48

efforts,3 nationally representative household surveys usually do not include specialized migration 49

modules or specific information to assess migration patterns between rural and urban areas (de 50

Brauw et al., 2014). In this paper, we exploit a unique module of the 2003 Education et Bien-être 51

des Ménages au Sénégal (Education and Household Welfare in Senegal) survey, which was 52

specifically designed to understand migration decisions by asking retrospective questions to young 53

adults, aged 21 to 35 years. Using these household data, we employ a multinomial logit model to 54

empirically estimate whether young people decide to migrate to either rural or urban areas. In 55

addition to individual characteristics, such age, gender, and ethnicity, we include childhood 56

demographic characteristics, such as the number and gender of siblings, the role of the family’s 57

financial constraints measured by the asset index of the household when the child was 10 years of 58

age, parents’ education, and shocks, including the death of their father and/or mother. Furthermore, 59

we control for childhood residence characteristics such as access to education and health centers. 60

The remainder of this paper is organized as follows. In Section 2, we describe the 61

conceptual framework that guides our empirical approach. Section 3 describes the household 62

survey data, including a discussion on how we define and classify migrants and a description of 63

3 Some recent panel household surveys have tracked individuals and migrants, such as the Kagera Household Survey (Beegle et al., 2011), and the World Bank’s Living Standards and Measurement Study Surveys, among others.

5

the patterns on internal mobility. We also describe our empirical strategy in this section. Section 4 64

presents the econometric results from the multinomial models that explain the determinants of 65

migration. Finally, Section 5 presents the discussion and conclusions. 66

67

2. Conceptual Framework 68

Most of the migration literature indicates that migrants are primarily young people (Lloyd, 2005; 69

Young, 2013), who seek to diversify and expand their economic opportunities, especially in 70

developing country contexts (McKenzie, 2008). Multiple individual, household, and contextual 71

factors encourage youth to migrate internally in search of opportunity, which makes the migration 72

process complex and context-specific (Massey et al., 1997; Heckert, 2015). 73

In contrast to earlier economic models of migration that analyze an individual’s decision 74

to move as a function of their own expected net economic benefit, looking for opportunities to 75

improve their economic status (Harris and Todaro, 1970), a growing literature—the New 76

Economics of Labor Migration (NELM)—has modeled migration as both an individual and a 77

family decision, which not only maximizes income but also minimizes risks (Stark, 1991; Stark 78

and Bloom 1985; Taylor 2001). If migration is an investment decision whereby individuals incur 79

costs to generate higher incomes, youth have lower costs in moving and have higher lifetime 80

expected returns. This is not only based on their longer life expectancy, compared to older people, 81

but also because the opportunity cost of young people in the place of origin can be lower due to, 82

for example, high youth unemployment rates. On the other hand, if migration is a family decision 83

and perceived as a risk-coping mechanism, the choice of which household member migrates is 84

based on both earning potential and the individual’s ability to be engaged in family insurance 85

arrangements. For instance, Rosenzweig and Stark (1989) show that Indian rural farm households 86

6

tend to engage in long-distance, marriage-cum-migration to cope with volatile profits. Also, 87

households might send young members to migrate, with the expectation that they will send 88

remittances back home (Heckert, 2015). 89

In this paper, we test the hypothesis of whether the decision to migrate is influenced by 90

individual characteristics as well as the circumstances in the households and communities where 91

young adults grew up, and whether these factors are differentiated by gender. Although we mostly 92

follow the NEML conceptual framework, which explains migration behavior by focusing on the 93

households’ characteristics in a broader societal context (Taylor and Martin 2001; De Haas 2010, 94

Tegegne and Desta, 2016), we build on the work of Abramitzky et al. (2013), who underscores the 95

role of childhood conditions on later migration decisions. Using a novel data set of the age of Mass 96

Migration (1850-1913) from Norway to the United States, Abramitzky et al. (2013) find evidence 97

that economic and family conditions of an individual’s household during childhood, particularly 98

parental wealth and gender composition of siblings, can shape the internal and international 99

migration decisions later in adult life. While some studies have analyzed the effect of individual 100

and household conditions, such as birth order and family size, on later economic outcomes such 101

as labor market performance (Psacharopoulos and Patrinos, 1997; Edmonds, 2006), there is little 102

evidence on how these conditions affect later internal migration decisions, and even less in the 103

context of developing countries. 104

Socioeconomic conditions during childhood, such as wealth and parents’ levels of 105

education can shape youth migration; nevertheless, it remains an empirical question as to how, and 106

in what direction, these factors affect internal migration flows. On the one hand, we can expect 107

that better-off households will be less likely to encourage their children to migrate, since the higher 108

their assets, the better the potential economic opportunities within the community in which the 109

7

young adults reside as a children.4 On the other hand, we can expect that asset-poor households 110

are less able to finance the costs of migration, and thus, their members are less likely to migrate.5 111

Indeed, McKenzie et al. (2007) show that the probability of migrating from Mexico to the United 112

States has an inverse U-shaped relationship with wealth. This nonlinear effect is explained by the 113

heterogeneity of migration networks: in sending communities with smaller migration networks, 114

the costs of migrating are relatively high, and wealth is positively correlated with the likelihood to 115

migrate; once the migration networks are larger, the costs, and thus the importance of wealth on 116

the decision to migrate, decreases. 117

Along the same lines, if migration is considered a family decision, the education of the 118

father and mother are expected to influence a young person’s decision to migrate (Smith and 119

Thomas, 1998; Quisumbing and McNiven, 2006). Parents’ education can be a proxy for other 120

household assets, such as networks and family connections, that can increase the probability of 121

migrating. Although we would expect that the more educated the parents, the more information is 122

available about the net benefits of migration, thereby increasing the odds of leaving, the empirical 123

evidence is not conclusive on the direction of the effect of parents’ education on migration of 124

family members (Pessino, 1991; Ezra et al., 2001). 125

Gender dynamics may also dictate whether youth migrate, their destination, and the extent 126

to which households invest in such decisions. There are reasons to believe that the drivers of 127

migration are different between women and men. Some empirical studies in developing countries 128

have shown that young women, unlike men, frequently move to marry (Smith and Thomas, 1998; 129

4 The land tenure systems in developing countries can affect the relationship between wealth and migration and, thus, shape youth migration decisions. For example, in the Philippines, young adults stay with their parents if they inherit land (Quisimbuing and McNiven, 2006). 5 Mendola (2008) that poorer households in Bangladesh are only able to afford domestic migration while the better-off households can afford the costs of international migration.

8

Reed, 2010; Chort et al., 2017). Also, gender differences are expected when parents encourage 130

daughters, rather than sons, to migrate because of the expectation that the former are more likely 131

to remit (World Bank, 2007). It is also possible that parents provide less financial support to their 132

daughters than their sons, because the parents internalize that their daughters’ migration returns 133

are lower than those of their sons (Heckert, 2015). 134

Furthermore, in Senegal, ethnicity plays an important role in female internal migration 135

(Brockerhoff et al.1993; Chort et al. 2017). Indeed, studies have shown that women’s internal 136

migration patterns may be related to the different marital and cultural traditions across ethnic 137

groups (Brockerhoff et al., 1993). For instance, Serere, Diola, and to a lesser extent, Wolof (Oulof) 138

women are more likely to migrate for reasons related to work opportunities than are Toucouleur, 139

Peul, or Soninke women, who virtually never migrate except with their spouses or families (Sy, 140

1991). More broadly, recent evidence from developing countries shows that different ethnic groups 141

can have different preferences toward migration related to, for example, historical shocks, 142

geographical situations, and ethnic-specific languages, among other factors. Thus, these 143

differences can lead that some ethnic groups encourage mobility from the village of birth or origin 144

while other groups can deter such movements. These different ethnic preferences can be shared 145

through social norms, and therefore, are likely to affect the decision-making of the individuals 146

within the group (Auwalin, 2019). Therefore, we account for ethnicity as a factor that can 147

contribute to internal migration. 148

Gender can also shape migration decisions through issues related to birth order and norms 149

regarding division of household roles and time use, which can include the division of household 150

work and labor market activities, or even marriage practices and cultural norms that shape an 151

individual’s migration decision. For instance, in the context of the migration from Norway to the 152

9

United States in the early 19th century, Abramitzky et al. (2013) show that men who had fewer 153

brothers and were the oldest in their families were less likely to migrate later in life, because the 154

eldest brother was the primary recipient of family inheritance. Younger brothers, having less 155

access to family resources, were more likely to migrate in search of better opportunities. In addition 156

to the household allocation of resources among siblings, there may also be a role played by rights 157

and tasks that relate to a child’s birth order position relative to their siblings. For example, Protik 158

and Kuhn (2007) show that, for Bangladesh, the migration of older brothers decreases the 159

likelihood of sisters to marry and reside in places far from their parents. One explanation the 160

authors give is that, in order to ensure elderly care be provided by their daughters, parents might 161

prevent a marriage that involved migration. Furthermore, there could be a substitution of tasks 162

among siblings of the same gender that shapes migration choices. For example, younger sisters are 163

less likely to migrate, since they assume expanded responsibilities for performing household 164

chores when replacing older siblings, who have previously migrated (Smith and Thomas, 1998; 165

Quisibuing and McNiven, 2006). 166

Although the NELM conceptual framework focuses on household determinants of 167

migration, most of the movements of youth from rural to urban areas is driven by the unequal 168

distribution of opportunities between these two areas (McKenzie, 2008). Opportunities available 169

to youth migrants depend on the social and economic characteristics in the migrants’ places of 170

origin (Heckert, 2015). Thus, our models account for whether the availability of community-level 171

social services during childhood can shape later-life migration decisions. Since public policy 172

determines the geographic distribution and disparity of social infrastructure, these variables help 173

us understand the role of government investments in migration choices. 174

10

Following this conceptual framework, we model young people’s decisions to migrate to 175

either rural or urban areas in Senegal as a function of their individual characteristics and their 176

childhood family and community circumstances prior to their departure. Our paper contributes to 177

the literature that explores the determinants and patterns of internal youth migration in developing 178

countries (Clark and Cotton, 2013; Beegle and Poulin, 2015; Heckert, 2015) by highlighting the 179

relevance of family- and community-level characteristics during childhood in predicting later 180

migration in life (Abramitzky et al., 2013). This analysis also contributes to the scant empirical 181

evidence on the determinants of female internal migration in developing countries (Assaad and 182

Arntz, 2005; Chort et al., 2017). 183

184

3. Data and Methods 185

3. 1 Data Sources and Descriptives 186

The data we use in this paper comes from the 2003 Household Survey on Education and Welfare 187

in Senegal (EMBS). From 28 rural and 32 urban communities (communes), 1,820 households were 188

surveyed. 6 , 7 The 2003 EBMS revisited children originally included in a 1995–96 survey: a 189

nationally representative, school-based survey known as PASEC (Programme d’Analyse des 190

Systemes Educatifs de la CONFEMEN). The PASEC survey administered tests of ability to a 191

sample of students (20 per school) in 2nd grade (CONFEMEN, 1999). The original PASEC cohort 192

6 EMBS was collected by the Centre de Recherches Economiques Appliquées (CREA), l’Université Cheikh Anta Diop (Senegal) and Cornell University. 7 Our household survey defines the rural and urban areas following the official definition by the Government of Senegal, specifically the Agence National de la Statistique et de la Demographie, which designates certain administrative areas as a commune de ville or urban area. Thus, urban areas consist of localities erected in communes regardless of the number of inhabitants, while rural areas (communautés rurales) correspond to the rest of the territory (ILO, 2018). It is worth noting that a commune is the smallest administrative level in Senegal. This definition has been valid since 1976; therefore, it is consistent throughout the period of our analysis and does not affect our results.

11

was not a representative sample of all children in the country, because it was school-based; thus, 193

it excluded children who had never enrolled, or who dropped out of school during their first year 194

of enrollment. To address the selection problem of excluding non-enrollees, in 2003 we 195

enumerated all the children and their households in the 60 original PASEC communities included 196

in our survey. We then randomly selected households with children of similar ages as those 197

children included in the original 60 PASEC communities. The participants in the 2003 survey thus 198

included those who were originally part of the PASEC sample and those that were not because 199

they were not enrolled in school at the time of the PASEC survey, either due to delayed enrollment 200

or because they never entered school. 201

As discussed by Glick and Sahn (2009, 2010), despite these efforts to address the selection 202

problem of enrollment, the sample is not truly nationally representative since it is part of a cohort 203

study of young children. Any cohort study will lose its representativeness over time. To mitigate 204

this concern, as discussed above, we randomly selected into the sample new households and their 205

children to ensure that the sample is as close as possible to a random sample of the villages that 206

were initially randomly selected from throughout the country. Of course, the problem remains that 207

the selection of villages sampled in 2003 was based on a listing from eight years earlier, that is, 208

there may be new villages that were formed between 1995 and 2003, which would not be included 209

in the sample. Considering these concerns, we made a comparison of descriptive statistics from 210

the survey with other national surveys. This effort was quite encouraging, since it showed that for 211

a range of demographic characteristics, as well as other characteristics such as education, the 212

EBMS sample of 1,820 households is consistent with those of a nationally representative sample.8 213

Likewise, the characteristics of the EBMS population, in terms of religion and ethnicity, are also 214

8 For example, the net primary enrollment in our sample (primary enrollments of children 7–12) is 66 percent, compared with 63 percent for the country as whole in 2000 (World Bank, 2006).

12

reflective of the nation as a whole. One small difference is that the proportion of rural households 215

in the 2003 EMBS is 53.2%, which is close to, but smaller than the rural population at the national 216

level of 59% according to the 2002 Census. 217

In our analysis, we rely extensively on the migration module of the EMBS, which contains 218

information on the current residence, the birthplace, and the residence five years prior to the survey 219

(1998). It also provides the years of residence in the current location. In addition, this module has 220

retrospective questions for adults above the age of 21 (migrants and nonmigrants) about where 221

they lived, as well as the household and community characteristics when they were 10 years old. 222

These data are key components of our methodology, because we can observe the childhood 223

characteristics of both migrants and nonmigrants that we use to analyze migration decisions. 224

Defining a migrant in empirical work is not always straightforward and often made difficult 225

due to limitations of the available data. We define migrants as individuals who have lived outside 226

of their communities for at least one year, departing from their place of origin after they were 10 227

years old.9 Among our sample of 2,676 individuals who fall in the age group of 21 to 35 years old, 228

35% are defined as migrants; in other words, 937 individuals left their communities for at least for 229

one year after they were 10 years old. It is worth noting that we are accounting for the last move 230

prior to the individual being surveyed, and as such we calculate the age of departure by subtracting 231

the number of years of residence in the destination (current place) from the young migrant’s current 232

age.10The median age of departure among these young migrants is 20 years. 233

9This definition is similar to Heckert (2015) who, in the context of Haiti, defines a migrant as an individual whose departure is after of 10 years old and has been outside from the place of origin at least for three years. 10 In other words, this “age of departure” is the age of arrival in the last residence. Although it is reasonable to assume only one migration experience at these young ages, this approach does not account for the possibility that there might be more than one migration experience.

13

We use the age range, 21–35 years old, because previous studies of internal migration have 234

shown that internal flows are the highest for individuals in this age group, especially as they search 235

for employment and better economic prospects (Brockerhoff et al., 1993; Ezra et al., 2001). This 236

cohort is especially important in terms of their experiences and recentness of their moves.11 237

We also suspect that the recall data is more accurate for these younger adults than for older 238

individuals. Furthermore, we test whether our results change if we exclude the individuals who 239

migrated at younger ages, between 10 and 14 years old who represent 15% of the sample migrants. 240

It is plausible that for these individuals, parents might strongly influence or make their decisions 241

to migrate. If this is the case, the migration decision will be endogenous to other household-level 242

decisions, such as fertility. We find that our key results are not sensitive to the choice of including 243

these younger migrants (see Table A.2 in the Appendix). 244

Although most of the empirical studies of internal migration in developing countries have 245

focused on out-migration, especially from rural areas, they have neglected a careful examination 246

of different patterns or types of migration such as rural-to-rural or sequential migration. Mainly, 247

this omission has been justified by the lack of data, as documented in the case of West Africa by 248

Beauchemin and Bocquier (2004). Among the few studies in developing countries, Pessino (1991) 249

analyzed the determinants of different types of migration in Peru. Identifying the movements by 250

the degree of urbanization of the origin, the author finds that primary migrants, that is, people who 251

move for the first time, are more likely to come from rural areas whereas repeat or return migrants, 252

that is, those people who have made prior moves, come from urban areas. Reed et al. (2010), using 253

11 To compare this number of internal migrants with other data sources in Senegal, we use the 2002 census and define an internal migrant as an individual that lives in a different region than the region of birth. We find that 21.65% of individuals, aged between 21 and 35, are internal migrants. Although this definition is different from the one used in this paper, the magnitude is comparable, as it does not include people that migrate and return within a shorter period of time, that is, our temporary migrants.

14

a household survey in Ghana, find that past and future mobility are positively and strongly 254

correlated, suggesting that previous mobility reduces the perceived cost of moving again. Another 255

important study that attempts to classify migrants is that of Juan and Kim (1979) who used census 256

data in the Philippines. The authors construct a comprehensive set of categories of migrants that 257

distinguishes migrants by various characteristics, including the number of moves and whether they 258

return to their birthplaces. 259

Building upon this previous work, and using the information from our survey on the place 260

of residence: (1) at the time of the survey (2003); (2) five years prior to the survey (1998); (3) 261

when individuals were 10 years old; and (4) when individuals were born, we first focus on the 262

periodicity of movements—that is, how many times the individual moves across these points in 263

time. We distinguish between primary (one move) and repeat migrants (two or more moves), as 264

well as return migrants. The latter category includes those whose second or third move involved 265

returning to their birthplace. To be included in the category of return migrants, by definition, they 266

have to report having lived at a location other than their birthplace either when they were 10 years 267

of age, in 1998, and/or at the time of the survey. In our sample, 25.4% are primary migrants, 3.0% 268

are secondary or tertiary migrants, and 11.9% are return migrants. A final and the largest group of 269

migrants—fully 59.6%—are those who we define as “temporary” migrants, but for whom we do 270

not have information on their migration, other than they were away from their birthplace for at 271

least one year. Thus, these individuals report that they were both residents in another location for 272

at least one year, but also that their birthplace is the same as their residence at the time of the 273

survey, and that they lived in their birthplace in 1998 as well as when they were 10 years old.12 274

12 Juan and Kim (1979) (as explained in Bilsborrow (1984)) classify these persons as nonmigrants, because they report the same place of residence at all points of time that are included in the survey. We acknowledge that there may be some misreporting among this group —that is, that they made an error in reporting having lived elsewhere for more than one year. However, we expect that the vast majority answered that question correctly and are indeed return

15

Table 1 shows the distribution of migration by the urban/rural origin and destination of the 275

move, as well as the migration categories: primary, return, repeat, and temporary, discussed 276

previously. We find that two-thirds of the migrants moving from rural to urban areas are primary 277

migrants; this is consistent with the fact that most of the migrants in Dakar are more likely to be 278

permanent migrants (World Bank, 2006). Interestingly, we also find that almost 60% of the urban-279

to-rural flows are of primary migrants. On the other hand, almost 60% of the rural-to-rural and 280

urban-to-urban migrants are temporary movers. Although our data do not allow us to capture trends 281

in migration, the descriptive statistics in Table 1 are consistent with other empirical evidence that 282

points out that rural-to-rural flows, and even the reverse urban-to-rural flows, have gained 283

prominence as internal migration movements in West Africa (Beauchemin and Bocquier, 2004; 284

Beauchemin, 2011). 285

286

<Insert Table 1 approximately here> 287

288

Table 2 summarizes the main socioeconomic characteristics of our sample. We include 289

temporary migrants in this table, and in the analysis that follows. Given that temporary migrants 290

can have different triggers to migrate internally from the rest of the migrants in the sample, we 291

tested that our main results are robust to excluding this group of temporary migrants from the 292

analysis (see Table A.3 in the Appendix). 293

Table 2 shows that our young migrants are mostly female. Women represent more than 294

two-thirds of the young migrants, compared to 53% in the nonmigrant group and 57% in the total 295

migrants, who happened not to live away from their place of birth in 1998 and when they were 10 years of age. In our analysis, we explore whether the results are sensitive to the inclusion or exclusion of these groups being characterized as migrants.

16

sample. The female overrepresentation in the group of young migrants can be explained 296

presumably by the association of migration and the decision to marry, as we will discuss further 297

in the next section. 298

299

<Insert Table 2 approximately here> 300

301

Our sample individuals have completed 4.3 years of schooling on average. Although the 302

school attainment is slightly higher for the nonmigrant group, compared with the migrant group, 303

this difference is statistically significant. We observe similar patterns regarding parents’ education: 304

more than 70% of the migrants’ fathers and 85% of their mothers did not go to school. Although 305

this situation is not appreciably different for nonmigrant young adults—68% of their fathers and 306

83% of their mothers did not go to school—the differences between these two groups are still 307

statistically significant. 308

Descriptive statistics on the access to social infrastructure when young migrants and non-309

migrants were 10 years old indicate that migrants come from areas with less access to a nearby 310

primary school, to a secondary school, and to a hospital.13 Approximately 91% of the young people 311

had a primary school near their residence. However, this percentage is only 86% for the young 312

migrants. Similarly, 45% of young migrants came from a community with a secondary school 313

nearby while this percentage was almost 10 points higher for the nonmigrants. Access to health 314

services was also unequal between migrants and nonmigrants in their childhood residences. While 315

71% of the migrants had access to hospitals, this percentage was 83% for the nonmigrant 316

population. 317

13 We define secondary school access as the existence of the school within 5 kilometers of a lower- or upper-level secondary school.

17

As noted earlier, we create an asset index following standard procedures, using factor 318

analysis and the dwelling characteristics where the young adults lived at 10 years of age.14 While 319

40% of the migrant children came from the lowest quartile, this percentage was 31% among the 320

nonmigrant group. However, this difference seems to be smaller for the highest quartile. Overall, 321

we find that the nonmigrant’s asset distribution, first order dominates that of the migrants. 322

323

3.2 Empirical Strategy 324

Empirical studies addressing the determinants of migration face the challenge of observing the 325

individual’s migration at one point in time after this decision has been made. Furthermore, the 326

decision to migrate can be made jointly with other household decisions, such as investments in 327

education and resource allocation, raising potential problems of endogeneity between migration 328

and its determinants. In a regression model, endogeneity is defined as a situation in which the 329

residual or error term is not statistically independent from one or more covariates (Wooldridge, 330

2002). This issue can occur when there is potential reciprocal or simultaneous causation between 331

the dependent and independent variables in the regression model. To a certain extent, and 332

following other demographic research (for example, Robles and Oropresa, 2011), we address this 333

issue by using a survey that includes retrospective data on young migrants and nonmigrants aged 334

21 to 35. This retrospective information on household and community characteristics of 335

individuals, when they were 10 years old, allows us to estimate the impact of childhood 336

circumstances long before they migrate, thereby reducing concerns over simultaneous causation 337

or reverse causality. Nevertheless, we acknowledge that we are not able to strictly establish 338

14 We construct the asset index based on the floor material, the source of potable water, and the type of bathroom for the dwelling. These were the only characteristics available in the retrospective survey module.

18

causality of the migration determinants, but rather explain whether these childhood determinants 339

are associated with migration among young adults. 340

Following our conceptual framework, the decision to migrate and where to migrate are 341

jointly modeled using a multinomial logit model in which individuals can decide between staying 342

(not moving), migrating to a rural area, or migrating to an urban area. We empirically test whether 343

the decision to migrate is influenced by individual, household, or community characteristics and 344

circumstances of their origins—that is, where the migrants grew up. These characteristics and 345

circumstances are based on those that existed when the individuals were 10 years old. More 346

specifically, we estimate the following reduced form regression equation:15 347

348

𝐿𝐿𝐿𝐿 �𝑝𝑝�𝑀𝑀𝑖𝑖

𝑘𝑘𝑘𝑘=1,2�

𝑝𝑝�𝑀𝑀𝑖𝑖𝑘𝑘𝑘𝑘=0�

� = 𝛼𝛼 + 𝑋𝑋𝑖𝑖𝛽𝛽𝑘𝑘 + 𝐸𝐸𝑖𝑖𝛿𝛿𝑘𝑘 + 𝐻𝐻𝑖𝑖𝜃𝜃𝑘𝑘 + 𝐶𝐶𝑖𝑖 𝜌𝜌𝑘𝑘 + 𝑅𝑅𝑖𝑖𝜋𝜋𝑘𝑘 + 𝜖𝜖𝜀𝜀𝑖𝑖𝑘𝑘 , 349

350

where 𝑀𝑀𝑖𝑖𝑘𝑘is the destination variable of individual i, k takes the value of 0 if the individual does not 351

migrate (the base case scenario), 1 if the individual migrates to a rural area, and 2 if the individual 352

migrates to an urban area. Xi represents individual characteristics such as age, ethnicity, and 353

gender.16 It is worth noting that in addition to running the models with a gender dummy variable, 354

we also account for differences in the determinants of migration by estimating separate models for 355

young women and men. We also control for Ei, the education of the individual’s parents. We 356

exclude from the models any current individual’s educational attainment because of its potential 357

15 Given that the independent variables are from the individual and not the destination choice, we are not required to implement a test of independence of irrelevant assumptions (IIA). 16 To mitigate concerns related to potential multicollinearity between ethnicity and other control variables, we have calculated the variance inflation factor (VIF), and it is less than 10, suggesting that this issue is not a concern. Our results are also robust to the exclusion of ethnicity as a control variable.

19

reversal causality with migration.17 Nevertheless, our results are qualitatively similar when we 358

include the individual’s years of education in our models (see Table A.1 in the Appendix). 359

Hi represents the household characteristics when the individuals were 10 years old. To 360

measure the household’s wealth and risk aversion, we include an asset index; as described earlier, 361

the index constructed was based on the dwelling conditions at age 10.18 We also include the 362

number and gender composition of the individual’s siblings, while acknowledging that these 363

variables can be in part a function of household preferences for the quality and quantity of children. 364

Nonetheless, the question of whether the presence of younger or older male and female siblings 365

contributes to migration provides for interesting insights about these relationships, even if we 366

cannot draw strict causal inferences from the results. We also control by whether either one or both 367

parents had passed away by the time the individual was 10 years old. We capture this by including 368

a dummy variable that takes the value of 1 when the individual reports that their father, mother, or 369

both passed away by the time they were 10 years old.19 370

Finally, Ci represents the community-level characteristics when individuals were 10 years 371

old. We include dummy variables for the access to primary and secondary schools and to hospitals 372

when the young adults were 10 years old. For each one of these variables, access is defined as the 373

existence of the corresponding institution within 5 kilometers from the individual’s residence when 374

they were 10 years old. Finally, we include Ri, a set of regional dummies corresponding to the 375

region of childhood residence, to control for social and economic characteristics that influence the 376

17 Using the 2003 EMBS data, we are not able to instrument the individual’s education at the time of the survey, nor can we infer the education completed before the migration decision. 18 In our models, we tested for an inverse U-shaped relationship between the asset index and the probability of migration by introducing a quadratic term in our regressions, but we did not find any statistically significant result for this nonlinearity. 19 We could not try a separate dummy variable for each parent’s death since the number of cases for either mother or father was too small.

20

costs of migration (for example, the distance to the capital, Dakar) that vary across regions, but 377

not over time.20 378

379

4. Results 380

Table 3 presents the average marginal effects of our multinomial models. Panel A shows the 381

average marginal effects for all the individuals between 21 and 35 years, while Panels B and C 382

show the results for young men and women, respectively. 383

384

<Insert Table 3 approximately here> 385

386

4.1 Individual Characteristics 387

From the model that includes both men and women, the negative and significant gender variable 388

indicates that women are 7.2% more likely than men to move to rural areas, although no gender 389

difference exists for moves to urban areas. These results may reflect that young women often move 390

as a consequence of following their spouses. While we are unable to prove the causal effect of 391

marriage on female youth migration, we examined the relationship between the age of marriage 392

and the age of migration. First, we note that, on average, among married couples, men are 12 years 393

older.21 Second, we notice that 72 percent of the women who migrate were already married, in 394

contrast to only 31 percent of the male migrants. These descriptive findings are consistent with 395

empirical evidence in Senegal showing that typically marriage is the main reason for migration 396

20 Our sample size is too small to accurately test the determinants of our models for each of the migration dyads: (1) rural-to-rural; (2) rural-to-urban; (3) urban-to-urban; and (4) urban-to-rural. 21 In the 2003 EMBS sample of married couples, the average woman’s age is 38 while for men, it is 50 years old.

21

among women of reproductive age (Safir, 2009), and that short-distance rural-to-rural marriage-397

related migrations are more frequent among women than men (Chort et al., 2017). 398

We also examine the marginal effect of age among the cohort of individuals between 21 399

and 35 years old, as shown in Panel A: being one year older increases the probability of migrating 400

to rural areas by 8.5% and decreases the probability of migrating to urban areas by 5%. While age 401

has no effect for men on the likelihood to migrate to urban areas, for women this effect varies with 402

their destination. As age increases, women are 10% more likely to migrate to rural areas and 6% 403

less likely to migrate to urban areas; however, this effect is nonlinear, as seen by the statistical 404

significance of the quadratic term, which indicates that the effect of age is not monotonic along 405

the age range of the women in our sample. 406

The results also show evidence that ethnicity influences the likelihood of migrating to rural 407

and urban areas.22 This effect is differentiated by gender. On the one hand, belonging to the Serere 408

group, relative to the Mendingue/Sose group that was excluded, decreases the likelihood of 409

migrating to urban areas by 17%. This marginal effect has a similar magnitude among women and 410

men. On the other hand, belonging to the Wolof group decreases only male migration to urban 411

areas by 11%, while belonging to the Poular group decreases only female migration to rural areas 412

by 8%. These results are in line with ethnographic evidence underlying the association between 413

ethnicity and migration, particularly for women, in West Africa (Bockefort et al., 1993). 414

415

4.2 Demographic and Economic Household Characteristics 416

Our results indicate that the children of fathers with more education are less likely to move to rural 417

areas and more likely to move to urban areas. Mother’s education, however, is not statistically 418

22 In our models, we include a dummy variable for missing observations, given the substantial amount of misreporting of this variable in the sample (523 observations for nonmigrants and 253 for migrants).

22

significant in any of our models.23 When examining the gender-disaggregated results, we observe 419

that the effect of the fathers’ education on youth migration is larger and statistically more robust 420

for their daughters than it is for their sons.24 This result may reflect the role of fathers in arranged 421

marriages, or perhaps in terms of promoting more educational opportunities for their daughters, 422

which often requires migrating to urban areas. In fact, these two mechanisms may be related: 423

greater education of the fathers, whether it be through ability, economic well-being, or more 424

expansive social networks, may enable them to find more favorable husbands for their daughters 425

who will move with their husbands to the city in pursuit of greater opportunities, or similarly, to 426

improve educational opportunities for their daughters, which requires schooling in urban areas. In 427

contrast, a father’s education may discourage marriage arrangements in which daughters would 428

migrate to rural areas, where the returns to education are likely to be lower.25 429

Our models also suggest that better living conditions during childhood, measured by the 430

dwelling asset index, are associated with the higher likelihood of migrating to urban areas while 431

decreasing the likelihood of migrating to rural areas; however, the latter effect is not statistically 432

significant.26 The asset index does not have a differentiated effect by gender. The result might 433

suggest that young women and men who grew up in asset-poor households are less able to afford 434

the costs of migration to urban areas. We also test if there was a differentiated effect of the asset 435

index by rural or urban origin. A better-off asset position of the household in a rural origin 436

23 We corroborate these results by estimating the same multinomial models and instead of parents’ highest education, we include dummy variables for whether each of the parents has some level of education. Results are available upon request. 24 The effect of the father’s education on young males is significant only at 10%, and it is not robust to the specification of a father’s literacy dummy variable. 25 Some empirical studies in African countries have shown that father’s education increases the education of both boys’ and girls’ schooling rather than mother’s education (Tansel, 1997), and in some cases, paternal education can favor more girls’ than boys’ education (Glick and Sahn, 2000). 26 This result is consistent with the fact that the asset distribution for migrants going to urban areas first order dominates the migrants going to rural areas.

23

decreases the likelihood of migrating to either rural or urban areas. Interestingly, this effect is 437

statistically significant for men and not for women, suggesting that male migration might be 438

deterred by better economic opportunities in rural areas, which are most probably associated with 439

agricultural activities (see Table A.4 in the Appendix). 440

The multinomial regressions in Table 3 also include information on the demographic make-441

up of the households when individuals were 10 years old. The results indicate that a higher number 442

of younger siblings increases the probability of migrating to rural areas, while a higher number of 443

older siblings does not have any effect on the probability of migrating to either urban or rural areas. 444

Looking at the models by gender, the results show that the marginal effect of having younger 445

siblings is still statistically significant for women, and this effect is only positive and significant 446

in the case of women moving to rural areas. One possible explanation is that women with a higher 447

number of younger sisters are more likely to migrate, because their young female siblings act as 448

substitutes in home production (Smith and Thomas, 1998; Quisimbuing and McNiven, 2006). 449

Indeed, we further examine the sex and birth order composition of the siblings in the likelihood of 450

migration. We estimate the multinomial models, including younger and older brothers and sisters 451

(see Table 4). We find that having younger sisters increases the odds of moving to rural areas, and 452

this effect is significant for women but not for men. 453

454

<Insert Table 4 approximately here> 455

456

In addition, we account for whether the individual has lost either their father or mother to 457

death, or both parents, by the time they were 10 years old. Our results indicate that the marginal 458

effect of the loss of a parent during childhood increases by 7% the probability of migrating to rural 459

24

areas, but it does not affect the likelihood of moving to urban areas. By gender, we find that loss 460

of a parent only affects female and not male migration, and this effect is only significant for those 461

women going to rural areas. Young people who have lost one or both parents are also more likely 462

to migrate, presumably reflecting weaker ties to their childhood places of residence. 463

464

4.3 Community Characteristics 465

The availability of social infrastructure, such as schools and hospitals, in the community where the 466

individual lived as a child, influences the probability of moving. Access to a primary school within 467

5 kilometers decreases the likelihood of moving to urban areas by 17.5%, but it does not affect the 468

probability of moving to rural areas. This marginal effect is of a similar magnitude for both men 469

and women. We also investigate whether the nearest primary school has a differentiated effect on 470

the likelihood to migrate based on whether the individual lived in a rural or urban area as a child. 471

To do so, we estimate models that include an interaction between the urban dummy and the nearest 472

primary school. We find that proximity to primary school decreases the probability of migrating 473

to urban areas only if the early childhood residence is in a rural area. Results are available upon 474

request. 475

Access to secondary school does not affect the decision to migrate in the aggregate sample; 476

however, when we examine the gender-disaggregated models, we find that a secondary school 477

within 5 kilometers actually increases the female probability of migration to urban areas by 10%. 478

We expect that this effect is mediated by the fact that access to secondary schools exposes girls 479

and their families to the potential of greater opportunities associated with education and increases 480

their openness to migrate in search of opportunity, whether in the labor market or in search of more 481

education. Proximity to a nearby hospital decreases the odds of migrating to rural areas only, but 482

again, this is only the case for potential women migrants. 483

25

In general terms, our results indicate that better access to social infrastructure during 484

childhood, particularly to primary schools and hospitals, deters later youth migration, consistently 485

with other empirical evidence in developing countries (Katz, 2000). However, there are potential 486

countervailing forces that could contribute to better social infrastructure, thereby encouraging 487

migration: that is, easier access to schools can also trigger migration if individuals who accumulate 488

more human capital in the presence of nearby schools migrate to other places to look for higher 489

returns to their capital accumulation. In fact, we find that women with access to secondary school 490

when they are 10 years of age are likelier to migrate to urban areas. 491

Finally, the dummy variable for whether the childhood residence was either rural or urban 492

corroborates the migration patterns described earlier: when the childhood residence is rural, the 493

likelihood to migrate to other rural areas increases by 15%; similarly, when the residence is urban, 494

the likelihood to migrate to urban areas increases by 7%. 495

496

5. Discussion and Conclusions 497

Our goal in this paper is to highlight the importance and magnitude of internal migration in Senegal 498

and to analyze the socioeconomic determinants of the decisions of young people to migrate 499

internally. We also examine whether these factors differ by gender. We focus on the role of 500

household and community characteristics during childhood, in the years prior to the decision to 501

migrate, using household survey data from Senegal that include retrospective information from the 502

time when individuals were 10 years old. Our multinomial logit model allows for individuals, 503

between 21 and 35 years, to decide between not migrating, or moving to rural or urban areas in 504

Senegal. 505

26

We find that more than a third of the individuals in our sample are migrants, and their 506

median age of departure is 20. Furthermore, we find that more than half of the total internal youth 507

migration is temporary and rural-to-rural or urban-to-urban, in contrast with the more widely 508

studied rural-to-urban permanent migration. Indeed, this finding highlights prior evidence from 509

documenting the relevance of these mobility patterns in francophone West Africa (Beauchemin 510

and Bocquier, 2004; Beauchemin, 2011). 511

Our findings suggest that the determinants of internal migration in Senegal are 512

heterogeneous by gender and differ for those leaving their childhood residence for an urban or 513

rural destination. Similar to Chort et al. (2017), we find that Senegalese women are more likely to 514

migrate for reasons related to marriage, something that has been documented in other sub-Saharan 515

African countries (Kudo, 2015). We also find that childhood socioeconomic conditions, such as 516

father’s education, the demographic composition of the household, and access to educational 517

opportunities where individuals grew up, can shape later youth migration differently for women 518

and men. For example, fathers’ education has a particularly important role in women’s migration 519

choices: the more educated the father, the more (less) likely are the daughters to move to urban 520

(rural) areas. In our sample, 72 percent of the female migrants are married. This result could 521

suggest that father’s education is influential in marriage arrangements and in the probability that a 522

daughter will marry someone and leave the childhood residence with her new husband in search 523

of greater economic opportunity in urban areas. These results are similar to those found by 524

Quisimbuing and McNiven (2006) in the Philippines, where father’s education increases the 525

probability of a daughter moving from the village, and interestingly, mother’s education has the 526

opposite effect. However, this is only conjecture, as we do not have further information to 527

disentangle the role of marriage and economic opportunities in the decision to migrate. 528

27

Furthermore, our findings suggest that the presence of younger siblings during childhood 529

is associated with migration decisions. For instance, women with younger sisters (but not brothers) 530

are more likely to migrate, suggesting that younger female siblings act as substitutes in household 531

responsibilities. We also find that those who lived in households with a higher asset index, when 532

they were 10 years old, are more likely to migrate to urban areas. This may be because these young 533

women and men are able to finance the costs of migrating to urban areas and to reap the benefits 534

of better employment opportunities in the cities. 535

The characteristics of the community in which children reside also shape migration 536

decisions. Proximity to better social infrastructure during childhood, particularly primary schools 537

and hospitals, is generally associated with a lower probability of migrating. The one clear 538

exception is access to secondary schools, which in fact increases the probability of migration to 539

urban areas for young women. While proximity to secondary schools may mitigate the need to 540

migrate in search of more education, such accessibility is likely associated with higher schooling 541

attainment, especially for girls whose parents are more reluctant to send their daughters away to 542

boarding schools and/or reside with relatives in order to raise school attainment. These human 543

capital investments may subsequently encourage migration of young women to urban areas in 544

search of employment opportunities that utilize their human capital and education. Although we 545

are not able to test this empirically, it is plausible that access to secondary school is more relevant 546

for women than men, because education has a larger effect on female than male migration. Indeed, 547

Chort et al. (2017) show that years of schooling increases the likelihood of migrating to urban 548

areas, especially for women, suggesting that education can be a channel to promoting women’s 549

migration, independent of the usual reason of migrating for family and marriage reasons. 550

28

Our findings motivate further research on the expected consequences of internal youth 551

migration for individuals, their households, and their communities. Even though migration can 552

expand labor market opportunities, some research has pointed out that young people are vulnerable 553

to negative migration experiences (Tienda et al., 2008; Heckert, 2015). Furthermore, while young 554

migrants can provide benefits to their households by sending remittances, the high costs of 555

financing migration and family disruptions could also negatively affect those households. 27 556

Similarly, while remittances can improve the economic conditions of the communities of origin, 557

migration can also be detrimental if the young, educated people leave their communities (as with 558

“brain drain”). Whether the benefits outweigh the costs of migration on individuals, households, 559

and communities remains an empirical question and cannot be answered generally. However, 560

future research can build on our findings by collecting long-term, longitudinal data, before and 561

after migration, thus allowing researchers to track the welfare consequences of internal migration 562

of young individuals, their households, and communities. This research can identify patterns and 563

circumstances which may enable policymakers to intervene to ensure the benefits of migration 564

outweigh its possible negative consequences. 565

While there is still much to be learned about the internal migration of young people in 566

Senegal, and more generally, in other developing countries, the high degree of mobility and the 567

recognition of certain factors that contribute to these population movements is important 568

knowledge for policymakers, both in terms of affecting and planning for the widespread migration. 569

While there remain many questions about the determinants of migration and how to cope with the 570

stresses on communities and households affected by these population movements, there is every 571

reason to expect that they will only accelerate in years to come. Indeed, in a country such as 572

27 The literature on the effects of remittances on household welfare is vast in developing countries. For instance, see Binci and Gianelli (2018) for a review of the effects of remittances on education and child labor.

29

Senegal where the young population will have doubled by 2035, and more than half of the 573

population still lives in the rural areas, factors such as increasing land pressure, the adverse effects 574

of climate change, and rapid structural transformation to a more industrialized and service-oriented 575

economy can be expected to increase internal youth mobility in the country (de Brauw et al., 2014; 576

Ba et al., 2017). 577

Although our analysis sheds light on whether childhood conditions influence later youth 578

mobility, it does not establish causality between the socioeconomic factors when young migrants 579

were 10 years old and their later internal mobility decisions in Senegal. To provide such causal 580

empirical evidence, future research could leverage experimental methods, an emerging 581

methodology in migration research (McKenzie, 2015), to study specific policy instruments for 582

managing internal migration. 583

30

References

Abramitzky, R., Boustan, L. P., & Abramitzky, K. (2013). Have the poor always been less likely to migrate? Evidence from inheritance practices during the age of mass migration. Journal of Development Economics, 102, 2–14.

Assaad, R., & Arntz, M. (2005). Constrained geographical mobility and gendered labor market

outcomes under structural adjustment: Evidence from Egypt. World Development, 33(3), 431–454.

Auwalin, I. (2019). Ethnic identity and internal migration decision in Indonesia. Journal of

Ethnic and Migration Studies, 1-21. Beauchemin, C., & Bocquier, P. (2004). Migration and urbanisation in francophone West Africa:

An overview of the recent empirical evidence.Urban Studies, 41(11), 2245–2272. Beauchemin, C. (2011). Rural–urban migration in West Africa: Towards a reversal? Migration

trends and economic situation in Burkina Faso and Côte d'Ivoire. Population, Space and Place, 17(1), 47–72.

Ba C, B. J and Diop. D. (2017). Senegal: the fluidity of internal migration as an answer to local

constraints. In : Rural Africa in motion. Dynamics and drivers of migration South of the Sahara. Mercandalli Sara (ed.), Losch Bruno (ed.). Rome : FAO-CIRAD, pp. 32-33. ISBN 978-92-5-109974-2

Beegle, K., & Poulin, M. (2013). Migration and the transition to adulthood in contemporary

Malawi. Annals of the American Academy of Political and Social Science, 648(1), 38-51. Beegle, K., De Weerdt, J., & Dercon, S. (2011). Migration and economic mobility in Tanzania:

Evidence from a tracking survey. Review of Economics and Statistics, 93(3), 1010–1033. Bilsborrow, R. E., Oberai, A. S., & Standing, G. (1984). Migration Surveys in Low Income

Countries: Guidelines for Survey and Questionnaire Design. London: Croom Helm. Binci, M., & Giannelli, G. C. (2016). Internal versus international migration: Impacts of

remittances on child labor and schooling in Vietnam. International Migration Review. Brockerhoff, M., & Eu, H. (1993). Demographic and socioeconomic determinants of female

rural to urban migration in Sub-Saharan Africa. International Migration Review, 27(3), 557–577.

Chort, I. (2014). Mexican migrants to the US: What do unrealized migration intentions tell us

about gender inequalities? World Development, 59, 535–552. Chort, I., De Vreyer, P., & Zuber, T. (2017). Gendered internal migration patterns in Senegal.

Working Paper No. DT/2017/02, DIAL (Développement, Institutions et Mondialisation).

31

Clark, S., & Cotton, C. (2013). Transitions to adulthood in urban Kenya: A focus on adolescent migrants. Demographic Research, 28, 1053–1092.

de Brauw, A., Mueller, V., & Lee, H. L. (2014). The role of rural–urban migration in the

structural transformation of Sub-Saharan Africa. World Development, 63, 33–42. CONFEMEN. (1999). Les facteurs de l’efficacité de l’enseignement primaire: données et

résultats sur cinq pays d’Afrique et de l’Océan Indien. CD-ROM. De Haas, H. (2010). Migration and development: A theoretical perspective 1. International

migration review, 44(1), 227-264. Edmonds, E. V. (2006). Understanding sibling differences in child labor. Journal of Population

Economics, 19(4), 795-821. Ezra, M., & Kiros, G. E. (2001). Rural out‐migration in the drought prone areas of Ethiopia: A

Multilevel Analysis1. International Migration Review, 35(3), 749–771. Giles, J., & Mu, R. (2007). Elderly parent health and the migration decisions of adult children:

Evidence from rural China. Demography, 44(2), 265–288. Glick, P., & Sahn, D. E. (2000). Schooling of girls and boys in a West African country: the

effects of parental education, income, and household structure. Economics of education review, 19(1), 63-87.

Glick, P., & Sahn, D. E. (2009). Cognitive skills among children in Senegal: Disentangling the

roles of schooling and family background. Economics of Education Review, 28(2), 178–188

Glick, P., & Sahn, D. E. (2010). Early academic performance, grade repetition, and school

attainment in Senegal: A panel data analysis. World Bank Economic Review, 24(1), 93–120.

Harris, J. R., & Todaro, M. P. (1970). Migration, unemployment and development: A two-sector

analysis. American Economic Review, 60(1), 126–142. Heckert, J. (2015). New perspective on youth migration: Motives and family investment patterns.

Demographic Research, 33(27), 765–800. ILO (2018) Retrieved from https://www.ilo.org/wcmsp5/groups/public/---dgreports/--

stat/documents/genericdocument/wcms_389373.pdf Juan, M. L. D., & Kim, Y. (1970). Migration differentials by migration types in the Philippines: A

study of migration typology and differentials.

32

Katz, E. (2000). Individual, household and community-level determinants of migration in Ecuador: Are there gender differences? Paper prepared for the Annual Meeting of the Population Association of America, Los Angeles, March 23-25.

Kudo, Y. (2015). Female migration for marriage: Implications from the land reform in rural

Tanzania. World Development, 65, 41–61. Lloyd, C. B (2005) National Research Council, National Research Council, & Committee on Population. Growing up global: The changing transitions to adulthood in developing countries. National Academy Press. Mendola, M. (2008). Migration and technological change in rural households: Complements or

substitutes? Journal of Development Economics, 85(1), 150–175. McKenzie, D. J. (2008). A profile of the world's young developing country international

migrants. Population and Development Review, 34(1), 115-135. McKenzie, D. (2015). Learning about Migration Through Experiments. Migration: Economic

Change, Social Challenge, 68. McKenzie, D., & Rapoport, H. (2007). Network effects and the dynamics of migration and

inequality: Theory and evidence from Mexico. Journal of Development Economics, 84(1), 1–24.

Massey, D.S., Arango, J., Hugo, G., Kouaouci, A., Pellegrino, A., and Taylor, J.E.(1993). Theories of international migration: A review and appraisal. Population and Development Review 19(3): 431–466. doi:10.2307/2938462. Pessino, C. (1991). Sequential migration theory and evidence from Peru. Journal of Development

Economics, 36(1), 55–87. Protik, A., & Kuhn, R. (2006). “The hidden cost of migration: Effect of brother’s migration on

sister’s marriage outcomes in rural Bangladesh.” Unpublished manuscript, Department of Economics, Brown University.

Patrinos, H. A., & Psacharopoulos, G. (1997). Family size, schooling and child labor in Peru–An

empirical analysis. Journal of population economics, 10(4), 387-405. Quisumbing, A., and McNiven, S. (2006). Migration and the rural-urban continuum: Evidence

from Bukidnon, Philippines. Philippine Journal of Development XXXIII (1–2), 1–43. Reed, H. E., Andrzejewski, C. S., & White, M. J. (2010). Men’s and women’s migration in

coastal Ghana: An event history analysis. Demographic Research, 22(25), 771–812. Robles, V. F., & Oropesa, R. S. (2011). International migration and the education of children:

Evidence from Lima, Peru. Population research and policy review, 30(4), 591-618.

33

Rosenzweig, M. R., & Stark, O. (1989). Consumption smoothing, migration, and marriage: Evidence from rural India. Journal of Political Economy, 97(4): 905–926.

Safir, A. (2009) Who leaves, who moves in? The impact of positive and negative income shocks

on migration in Senegal. Working Paper No 76, World Bank, Washington, DC. Smith, J. P., & Thomas, D. (1998). On the road: Marriage and mobility in Malaysia. Journal of

Human Resources, 33(4), 805–832. Stark, O., & Bloom, D. E. (1985). The new economics of labor migration. American Economic

Review, 75(2), 173–178. Stark, O. (1991). The Migration of Labor (pp. 216–220). Oxford: Blackwell. Sy, M. (1991). Reasons for Senegalese migration determined by ethnic background and social

status. Pop Sahel, (16), 29-35. Tansel, A. (1997). Schooling attainment, parental education, and gender in Cote d'Ivoire and

Ghana. Economic Development and Cultural Change, 45(4), 825-856. Taylor, J. E. (1999).The new economics of labor migration and the role of remittances in the

migration process. International Migration 37(1), 63–88. __________ (2001). Migration: New dimensions and characteristics, causes, consequences and

implications for rural poverty. Food, agriculture, and rural development: Current and emerging issues for economic analysis and policy research, 167. http://www.fao.org/docrep/003/X9808e/x9808e07.htm#ee

Taylor, J. E., & Martin, P. L. (2001). Human capital: Migration and rural population change. In

Handbook of Agricultural Economics, Vol. 1, Part A, ed. B. L. Gardner and G. C. Rausser, 457-511. Amsterdam: Elsevier.

Tegegne, A. D., & Penker, M. (2016). Determinants of rural out-migration in Ethiopia: Who

stays and who goes?. Demographic Research, 35, 1011-1044. Tienda, M., Taylor, L., and Moghan, J. (2007). New frontiers, uncertain futures: Migrant youth

and children of migrants in a globalised world. Zürich, Switzerland: Background paper prepared for Zurich Migration Workshop (Vol.17), August 26 2007.

United Nations Development Program (UNDP) (2009). Overcoming Barriers: Human Mobility

and Development. Human Development Report. New York: Palgrave Macmillan. Wooldridge, J. M. (2002). Econometric analysis of cross section and panel data MIT

Press. Cambridge, MA, 108.

34

World Bank (2006). Senegal second Poverty Reduction Paper Strategy. World Bank, Washington, DC.

World Bank (2007). Development and the next generation. World Development Report, World

Bank. Washington, DC. World Bank (2019). World Development Indicators Young, A. (2013) 'Inequality, the Urban-Rural Gap, and Migration', The Quarterly Journal of Economics 128: 1727-1785.

35

TABLES

Table 1: Distribution of Migrants by Rural/Urban Birthplace and 2003 Residence

Urban–Rural Rural–Rural Urban–Urban Rural–Urban

Primary 60.6% 7.3% 26.2% 59.8% Repeat 3.0% 0.5% 4.3% 8.3% Return 1.5% 14.7% 14.5% 2.3% Temporary 34.8% 77.5% 55.1% 29.5% Total 100% 100% 100% 100%

N a 66 409 325 132 a Refers to the total number of migrants by urban–rural destination.

36

Table 2: Socioeconomic Characteristics of Migrants and Nonmigrants

Migrant Non-migrant Total

Individual Characteristics in 2003 Percentage female 64% 53% 57% Average age 27.79 26.40 26.90 (4.55) (4.42) (4.52) Years of education 4.14 4.45 4.34 (4.61) (4.25) (4.38) Ethnicity groups (%)*

Wolof 29.4% 35.8% 33.5% Poular 24.7% 20.0% 21.7% Sose 13.8% 17.84% 15% Serere 20.4% 16.2% 18.9% Diola 8.2% 5.0% 6.2%

% whose Father has no education 73.1% 69.2% 70.6% % whose Mother has no education 85.4% 82.7% 83.7% Characteristics at age of 10 years Average number of older siblings 1.80 1.88 1.85 (2.01) (2.05) (2.04) Average number of younger siblings 2.57 2.42 2.47 (2.09) (2.10) (2.10) Access to primary school 86% 95% 91% Access to secondary school 45% 55% 51% Access to hospital 71% 83% 79% Distribution by asset quartiles First 40.22% 31.5% 34.6% Second 19.57% 18.1% 18.6% Third 28.60% 25.6% 23.1% Fourth 21.61% 24.8% 23.7% N 855 1546 2401

Notes: Standard deviations in parenthesis. Other ethnicity and regional dummy variables are not shown. Individuals from other ethnicities represent 4% of the sample.

37

Table 3: Average Marginal Effects of Multinomial Logits by Rural and Urban Destination

Panel A Panel B Panel C ALL MEN WOMEN Rural Urban Rural Urban Rural Urban Individual Characteristics Gender –0.072*** –0.009 (0.015) (0.015) Age 0.085*** –0.052** 0.061* –0.028 0.096*** –0.058* (0.024) (0.024) (0.034) (0.037) (0.034) (0.031) Age-squared –0.001*** 0.001** –0.001+ 0.001 –0.002*** 0.001** (0.000) (0.000) (0.001) (0.001) (0.001) (0.001) Wolof –0.025 –0.049+ –0.026 –0.111** –0.029 –0.000 (0.033) (0.032) (0.047) (0.046) (0.046) (0.044) Poular –0.016 –0.021 0.045 –0.054 –0.079** 0.008 (0.027) (0.032) (0.034) (0.046) (0.040) (0.045) Serere 0.004 –0.167*** –0.030 –0.155*** 0.033 –0.174*** (0.035) (0.042) (0.048) (0.058) (0.050) (0.059) Diola –0.027 –0.006 0.020 –0.013 –0.056 –0.003 (0.042) (0.044) (0.055) (0.063) (0.059) (0.061) Other ethnicity –0.099* –0.071+ –0.062 –0.141* –0.131* –0.011 (0.053) (0.047) (0.083) (0.073) (0.071) (0.063) Household Characteristics Father’s education –0.018*** 0.012*** –0.017* 0.006 –0.022** 0.018*** (0.007) (0.004) (0.009) (0.006) (0.009) (0.005) Mother’s education –0.006 0.004 0.005 –0.006 –0.012 0.008 (0.010) (0.006) (0.013) (0.009) (0.014) (0.007) Asset index (z-score) –0.007 0.029** –0.021 0.033* –0.000 0.022 (0.012) (0.012) (0.018) (0.019) (0.017) (0.016) Older siblings –0.002 –0.002 –0.009+ –0.006 0.004 –0.000 (0.004) (0.004) (0.006) (0.005) (0.006) (0.005) Younger siblings 0.009** 0.000 0.002 0.005 0.013** –0.004 (0.004) (0.004) (0.005) (0.005) (0.005) (0.005) Loss of parent(s) 0.070*** 0.035 0.068* 0.062 0.077** 0.020 (0.027) (0.030) (0.042) (0.051) (0.036) (0.038) Community Characteristics Primary school –0.024 –0.175*** 0.027 –0.198*** –0.063* –0.170*** (0.024) (0.037) (0.034) (0.061) (0.033) (0.048) Secondary school –0.002 0.028 0.034 –0.054 –0.033 0.102** (0.026) (0.029) (0.034) (0.041) (0.037) (0.040) Hospital –0.074*** 0.044 –0.039+ 0.080+ –0.097*** 0.020 (0.020) (0.035) (0.027) (0.055) (0.028) (0.045) Rural 10 years 0.155*** –0.070** 0.168*** –0.130*** 0.129*** –0.022 (0.031) (0.029) (0.043) (0.043) (0.043) (0.040) Notes: *** p<0.01, ** p<0.05, * p<0.1, + p<0.15. Standard errors calculated using the delta method. All models include regional dummies for childhood place when 10 years old. Number of observations: ALL: 2,401; Men: 1,035; Women: 1,366.

38

Table 4: Average Marginal Effects including Siblings’ Gender and Age Composition

Panel A Panel B Panel C ALL MEN WOMEN Rural Urban Rural Urban Rural Urban Father’s education –0.018*** 0.012*** –0.017* 0.006 –0.023** 0.018*** (0.007) (0.004) (0.009) (0.006) (0.009) (0.005) Mother’s education –0.005 0.004 0.006 –0.006 –0.011 0.008 (0.010) (0.006) (0.013) (0.009) (0.014) (0.007) No. older brothers 0.003 0.006 –0.014 –0.000 0.017* 0.013+ (0.006) (0.006) (0.010) (0.008) (0.009) (0.008) No. older sisters –0.007 –0.013* –0.005 –0.013 –0.010 –0.015+ (0.007) (0.007) (0.011) (0.010) (0.010) (0.009) No. younger brothers 0.005 –0.002 –0.001 0.001 0.010 –0.007 (0.005) (0.006) (0.007) (0.008) (0.008) (0.008) No. younger sisters 0.013** 0.002 0.007 0.007 0.016** –0.002 (0.006) (0.006) (0.008) (0.008) (0.008) (0.008) Asset index (z-score) –0.007 0.029** –0.021 0.033* –0.000 0.022 (0.012) (0.012) (0.018) (0.019) (0.017) (0.016) Loss of parent(s) 0.070*** 0.038 0.064+ 0.064 0.078** 0.021 (0.027) (0.030) (0.042) (0.051) (0.036) (0.038) Notes: *** p<0.01, ** p<0.05, * p<0.1, + p<0.15. Standard errors calculated using the delta method. All models include individual and community variables as well as regional dummies for childhood place when 10 years old. Number of observations: ALL: 2,401; Men: 1,035; Women: 1,366.

39

ONLINE APPENDIX (NOT FOR PUBLICATION)

Table A.1: Average Marginal Effects—Main Results including Individual’s Education

Panel A Panel B Panel C ALL MEN WOMEN Rural Urban Rural Urban Rural Urban Individual Characteristics Gender –0.075*** –0.014 (0.016) (0.015) Age 0.088*** –0.051** 0.074** –0.034 0.096*** –0.054* (0.024) (0.024) (0.034) (0.037) (0.034) (0.031) Age-squared –0.001*** 0.001** –0.001* 0.001 –0.002*** 0.001* (0.000) (0.000) (0.001) (0.001) (0.001) (0.001) Years of education 0.002 0.003+ 0.008*** 0.005** –0.006* 0.001 (0.002) (0.002) (0.002) (0.003) (0.003) (0.003) Household Characteristics Father’s education –0.017** 0.010** –0.019** 0.003 –0.016* 0.017*** (0.007) (0.004) (0.009) (0.006) (0.010) (0.005) Mother’s education –0.011 0.004 0.004 –0.007 –0.020 0.009 (0.011) (0.006) (0.013) (0.009) (0.016) (0.008) Asset index (z-score) –0.007 0.026** –0.025 0.029+ 0.006 0.019 (0.012) (0.012) (0.018) (0.019) (0.017) (0.016) No. older siblings –0.001 –0.004 –0.011* –0.008 0.006 –0.001 (0.004) (0.004) (0.006) (0.006) (0.006) (0.005) No. younger siblings 0.008** 0.001 0.002 0.005 0.013** –0.003 (0.004) (0.004) (0.004) (0.005) (0.005) (0.005) Loss of parent(s) 0.067** 0.044+ 0.074* 0.075 0.067* 0.025 (0.027) (0.031) (0.042) (0.052) (0.036) (0.038) Community Characteristics Primary school –0.027 –0.174*** 0.014 –0.192*** –0.058* –0.167*** (0.024) (0.038) (0.033) (0.062) (0.034) (0.048) Secondary school –0.003 0.025 0.029 –0.049 –0.035 0.098** (0.026) (0.029) (0.033) (0.041) (0.038) (0.041) Hospital –0.073*** 0.044 –0.044+ 0.080+ –0.089*** 0.017 (0.020) (0.035) (0.027) (0.055) (0.028) (0.045) Rural at 10 years 0.159*** –0.074** 0.171*** –0.124*** 0.122*** –0.031

(0.031) (0.030) (0.042) (0.043) (0.044) (0.040) Notes: *** p<0.01, ** p<0.05, * p<0.1, + p<0.15. Standard errors calculated using the delta method. All models include regional dummies for childhood place at 10 years old. Number of observations: ALL: 2,401; Men: 1,035; Women: 1,366.

40

Table A.2: Average Marginal Effects—Main Results excluding the Youngest Migrants

ALL Rural Urban

Individual Characteristics Gender –0.071*** –0.012

(0.015) (0.015) Age 0.077*** –0.022

(0.024) (0.024) Age-squared –0.001*** 0.001

(0.000) (0.000) Household Characteristics

Father’s education –0.014** 0.010*** (0.006) (0.004)

Mother’s education –0.006 0.004 (0.010) (0.006)

Asset index (z_score) –0.004 0.028** (0.012) (0.012)

Older siblings –0.001 –0.001 (0.004) (0.004)

Younger siblings 0.008** 0.001 (0.003) (0.003)

Loss of parent(s) 0.063** 0.047* (0.027) (0.029)

Community Characteristics Primary school –0.029 –0.149***

(0.024) (0.037) Secondary school 0.006 0.017

(0.026) (0.028) Hospital –0.078*** 0.044

(0.020) (0.034) Rural at10 years old 0.156*** –0.065**

(0.031) (0.028) Notes: *** p<0.01, ** p<0.05, * p<0.1, + p<0.15. Standard errors calculated using the delta method. All models include regional dummies for childhood place when 10 years old. Number of 0bservations: ALL: 2,274.

41

Table A.3: Average Marginal Effects—Main Results excluding Temporary Migrants

ALL Rural Urban

Gender –0.026** –0.008 (0.012) (0.014)

Age 0.048** –0.022 (0.020) (0.023)

Age-squared –0.001** 0.000 (0.000) (0.000)

Household Characteristics Father’s education –0.007+ 0.009**

(0.004) (0.004) Mother’s education –0.002 0.001

(0.006) (0.006) Asset index –0.006 0.042***

(0.010) (0.012) No. older siblings –0.004 –0.002

(0.003) (0.003) No. younger siblings 0.001 –0.005+

(0.003) (0.004) Loss of parent(s) 0.037* 0.057**

(0.020) (0.027) Community Characteristics

Primary school –0.019 –0.165*** (0.022) (0.034)

Secondary school 0.012 0.030 (0.020) (0.027)

Hospital –0.013 0.027 (0.018) (0.032)

Rural at 10 years 0.054** –0.001 (0.024) (0.028)

Notes: *** p<0.01, ** p<0.05, * p<0.1, + p<0.15. Standard errors calculated using the delta method. All models include ethnicity dummies and regiona dummies for childhood place when 10 years old. Number of observations 1,897

42

Table A.4: Main Results including interaction between Asset Index and Rural Origin

Panel A Panel B Panel C

All MEN Women Rural Urban Rural Urban Rural Urban

Father’s education –0.018*** 0.012*** –0.016* 0.006 –0.022** 0.018*** (0.007) (0.004) (0.009) (0.006) (0.009) (0.005) Mother’s education –0.006 0.004 0.006 –0.006 –0.012 0.008 (0.010) (0.006) (0.012) (0.009) (0.014) (0.007) Older siblings –0.002 –0.002 –0.010* –0.006 0.004 –0.000 (0.004) (0.004) (0.006) (0.005) (0.006) (0.005) Younger siblings 0.009** 0.000 0.003 0.005 0.013** –0.004 (0.003) (0.004) (0.004) (0.005) (0.005) (0.005) Asset index 0.029+ 0.033** 0.032 0.044** 0.027 0.020 (0.020) (0.014) (0.032) (0.021) (0.027) (0.019) Rural at 10 years 0.148*** –0.080*** 0.165*** –0.164*** 0.126*** –0.022 (0.030) (0.031) (0.047) (0.049) (0.045) (0.040) Rural*asset –0.056** –0.022 –0.079** –0.068* –0.042 0.001 (0.024) (0.024) (0.038) (0.040) (0.033) (0.031) Loss of parent(s) 0.069*** 0.034 0.070* 0.059 0.075** 0.020 (0.027) (0.030) (0.041) (0.051) (0.036) (0.038) Notes: *** p<0.01, ** p<0.05, * p<0.1, + p<0.15. Standard errors calculated using the delta method. All models include individual and community variables as well as regional dummies for childhood place when 10 years old. Number of observations: ALL: 2,401; Men: 1,035; Women: 1,366.